A comparison of proxy variable and stochastic latent variable approaches to the measurement of bias in technological change in south african agriculture

Technical change is inherently unobservable and has conventionally been represented by proxy variables, from simple time trends to more sophisticated knowledge stock variables. This paper follows Lambert and Shonkwiler (1995) in modelling technical change as a stochastic unobservable variable and tests this formulation against the alternative of using R&D and patent indices. This is done by fitting a system of share equations, derived from the dual profit function, to production data for South African agriculture. Each equation includes both unobserved technical change components and technical proxy variables. Variable deletion tests show that conventional proxy variables fail to explain the biases of technological change, while cointegration tests show that technical change is both stochastic and biased. The latent variables provide estimates of biases that are consistent with past studies and the historical record and can be explained by policy change in South Africa following WWII. The demonstration of high rates of return to R&D is not sufficient to justify R&D activity when biased technological change exacerbates input use and welfare distortions within and without the sector. * We thank the University of Pretoria for funding the study and the referees and delegates for many useful comments.